apunkt/plano
1
0
Fork 0
mirror of https://github.com/katanemo/plano.git synced 2026-04-28 18:36:34 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update arch_config and add tests for arch config file (#407)

Browse source
This commit is contained in:
Adil Hafeez 2025-02-14 19:28:10 -08:00 • committed by GitHub
parent d0a783cca8
commit e40b13be05
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
31 changed files with 379 additions and 212 deletions
Download patch file Download diff file Expand all files Collapse all files

60
arch/envoy.template.yaml
View file

@ -29,11 +29,11 @@ stats_config:
- 180000
static_resources:
listeners:
- name: arch_listener_http
- name: ingress_traffic
address:
socket_address:
address: 0.0.0.0
port_value: 10000
address: {{ prompt_gateway_listener.address }}
port_value: {{ prompt_gateway_listener.port }}
traffic_direction: INBOUND
filter_chains:
- filters:
@ -55,7 +55,7 @@ static_resources:
random_sampling:
value: {{ arch_tracing.random_sampling }}
{% endif %}
stat_prefix: arch_listener_http
stat_prefix: ingress_traffic
codec_type: AUTO
scheme_header_transformation:
scheme_to_overwrite: https
@ -76,13 +76,13 @@ static_resources:
route:
auto_host_rewrite: true
cluster: arch_prompt_gateway_listener
timeout: 60s
timeout: {{ prompt_gateway_listener.timeout }}
http_filters:
- name: envoy.filters.http.router
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
- name: arch_prompt_gateway_listener
- name: ingress_traffic_prompt
address:
socket_address:
address: 0.0.0.0
@ -104,11 +104,11 @@ static_resources:
envoy_grpc:
cluster_name: opentelemetry_collector
timeout: 0.250s
service_name: prompt_processor
service_name: ingress_traffic
random_sampling:
value: {{ arch_tracing.random_sampling }}
{% endif %}
stat_prefix: arch_prompt_gateway_listener
stat_prefix: ingress_traffic
codec_type: AUTO
scheme_header_transformation:
scheme_to_overwrite: https
@ -201,7 +201,7 @@ static_resources:
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
- name: arch_internal
- name: egress_api_traffic
address:
socket_address:
address: 0.0.0.0
@ -223,11 +223,11 @@ static_resources:
envoy_grpc:
cluster_name: opentelemetry_collector
timeout: 0.250s
service_name: prompt_processor
service_name: egress_api_traffic
random_sampling:
value: {{ arch_tracing.random_sampling }}
{% endif %}
stat_prefix: arch_internal
stat_prefix: egress_api_traffic
codec_type: AUTO
scheme_header_transformation:
scheme_to_overwrite: https
@ -273,13 +273,12 @@ static_resources:
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
- name: arch_listener_http_llm
- name: egress_traffic
address:
socket_address:
address: 0.0.0.0
port_value: 12000
traffic_direction: INBOUND
address: {{ llm_gateway_listener.address }}
port_value: {{ llm_gateway_listener.port }}
traffic_direction: OUTBOUND
filter_chains:
- filters:
- name: envoy.filters.network.http_connection_manager
@ -300,7 +299,7 @@ static_resources:
random_sampling:
value: {{ arch_tracing.random_sampling }}
{% endif %}
stat_prefix: arch_listener_http
stat_prefix: egress_traffic
codec_type: AUTO
scheme_header_transformation:
scheme_to_overwrite: https
@ -321,14 +320,13 @@ static_resources:
route:
auto_host_rewrite: true
cluster: arch_listener_llm
timeout: 60s
timeout: {{ llm_gateway_listener.timeout }}
http_filters:
- name: envoy.filters.http.router
typed_config:
"@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router
- name: arch_listener_llm
- name: egress_traffic_llm
address:
socket_address:
address: 0.0.0.0
@ -349,11 +347,11 @@ static_resources:
envoy_grpc:
cluster_name: opentelemetry_collector
timeout: 0.250s
service_name: llm_gateway
service_name: egress_traffic_llm
random_sampling:
value: {{ arch_tracing.random_sampling }}
{% endif %}
stat_prefix: arch_listener_http
stat_prefix: egress_traffic
codec_type: AUTO
scheme_header_transformation:
scheme_to_overwrite: https
@ -443,7 +441,7 @@ static_resources:
clusters:
- name: openai
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -467,7 +465,7 @@ static_resources:
tls_minimum_protocol_version: TLSv1_2
tls_maximum_protocol_version: TLSv1_3
- name: mistral
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -488,7 +486,7 @@ static_resources:
sni: api.mistral.ai
{% for internal_cluster in ["arch_fc", "model_server"] %}
- name: {{ internal_cluster }}
connect_timeout: 5s
connect_timeout: 0.5s
type: STRICT_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -504,7 +502,7 @@ static_resources:
hostname: {{ internal_cluster }}
{% endfor %}
- name: mistral_7b_instruct
connect_timeout: 5s
connect_timeout: 0.5s
type: STRICT_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -523,7 +521,7 @@ static_resources:
{% if cluster.connect_timeout -%}
connect_timeout: {{ cluster.connect_timeout }}
{% else -%}
connect_timeout: 5s
connect_timeout: 0.5s
{% endif -%}
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
@ -557,7 +555,7 @@ static_resources:
{% for local_llm_provider in local_llms %}
- name: {{ local_llm_provider.name }}
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -589,7 +587,7 @@ static_resources:
{% endfor %}
- name: arch_internal
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -605,7 +603,7 @@ static_resources:
hostname: arch_internal
- name: arch_prompt_gateway_listener
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN
@ -621,7 +619,7 @@ static_resources:
hostname: arch_prompt_gateway_listener
- name: arch_listener_llm
connect_timeout: 5s
connect_timeout: 0.5s
type: LOGICAL_DNS
dns_lookup_family: V4_ONLY
lb_policy: ROUND_ROBIN